Shift to On-Premises AI Data Centers

0:00 these are our 2024 data center

0:02 predictions AI data centers to cloud or

0:05 not to Cloud AI version we've seen this

0:08 movie before 5 or 10 years ago

0:11 Enterprises started rushing to public

0:13 Cloud enticed by utopian Promises of

0:16 Greater flexibility and lower costs

0:19 however most companies eventually

0:20 realize that public cloud is not as

0:22 simple and cheap as so many had thought

0:25 Cloud regret is the new story as we hear

0:28 countless anecdotes of companies rria

0:30 workloads back to private on Prem data

0:32 centers we are now seeing the beginnings

0:35 of these same decisions with respect to

0:37 plans for new AI data center

0:39 infrastructure but this time companies

0:41 will be smarter and more thoughtful

0:44 build on Prem versus rent public Cloud

0:46 versus hybrid both all of these

0:48 decisions will be made more lucidly and

0:51 professionally yes building out a new

0:53 GPU cluster is expensive but so is

0:56 renting that capacity from a hyperscaler

0:59 many companies will opt to build on-prem

1:01 AI data centers for better control

1:04 tighter security and yes lower costs

1:06 than public Cloud Alternatives

1:08 infrastructure vendors will continue to

1:10 design and build more efficient gear but

1:13 driven by the rigorous demands of new AI

1:15 model training data center racks will

1:18 continue to consume more power from 10

1:20 Kow to now over 100 kilow for racks in

1:23 some places this places enormous demands

1:26 on data center facilities in number one

1:29 power needs and number two cooling

1:31 Renewables will rise in importance 100%

1:34 renewable powered data centers will be

1:36 typical rather than a rarity data center

1:39 build plans will increasingly be

1:41 conscious of geography cool climates

1:43 with access to Sun wind Hydro

1:46 experimental cooling methods that

1:48 minimize power consumption such as

1:50 liquid immersion will emerge from the

1:52 experimentation stages customers are

1:54 even being asked to byop bring their own

1:57 power to collocation facilities the the

1:59 cloud Engineers are the cool kids and

2:01 they keep moving in on the traditional

2:03 Network Engineers Turf while old school

2:06 Network Tools still survive you can just

2:08 never seem to get rid of Legacy Cloud

2:10 skills will increasingly be used to run

2:12 not just public clouds but also private

2:15 infrastructure Network Engineers don't

2:17 need to become software developers to

2:19 survive but they do need to be fluent in

2:22 Cloud tools such as terraform

2:24 traditional networking tools that can

2:26 seamlessly integrate with Cloud

2:27 technologies will Thrive while bulky

2:30 siloed old Network Management Systems

2:32 knots built for integration first will

2:35 wither with explosive growth in

2:37 applications and workloads delivering on

2:39 end user experience and requirements is

2:41 now more crucial than ever up to this

2:44 point AI Ops and networking centered

2:46 around security and simpler use cases in

2:49 campus and Branch environments the first

2:52 big use case area for AI Ops in the DC

2:55 will be predictive maintenance and

2:57 troubleshooting tools will look for

2:59 patterns that that typically foreshadow

3:01 problems and proactively notify it with

3:04 the changes needed to prevent

3:05 performance degradation or outages when

3:08 problems do occur AI Ops will actively

3:11 perform the troubleshooting steps

3:13 normally taken by a network operator and

3:15 present the results significantly

3:17 reducing meantime to repair and just as

3:20 importantly meantime to innocence AI

3:23 based large language models will be

3:25 incorporated in virtually every

3:26 interface enabling operators to rapidly

3:29 navigate complex systems to get answers

3:32 about everything from current Network

3:34 state to configuration changes or

3:36 recommended

3:47 [Music]

3:58 upgrades